How Is Agentic AI Reshaping Finance? Robinhood Continues Building While Regulators Race to Keep Up

Welcome to Agentic AI in Financial Services, the first newsletter dedicated to the transformation in financial services driven by the advancements in AI, shaping the Agentic economy and crafted by a team of seven AI agents that keep evolving and adapting.

These agents scan the world daily, filtering hundreds of news articles. We work together to bring you a weekly curated selection coupled with my human insights.

Thanks for reading and being part of our fast-growing subscription community, which just crossed 40,000 readers. Let's stay up to date and build a more human world powered by agentic AI.

This week, Robinhood used a London keynote to fuse Robinhood Chain the infrastructure for Tokenized stocks, Stock Tokens, and Agentic Accounts for Trading into one architecture play, capped with a Guinness World Record for an AI agent shopping autonomously on its Agentic Credit Card. Meanwhile, Kueski 's AI-first underwriting is proving out real financial inclusion in Mexico, and OCBC is betting on AI-native wealth advice and 600 new hires can co-exist in Singapore.

On the regulatory front, three separate institutions moved in the same week: the FCA and Bank of England called agentic AI a "profound step change" and signaled they won't wait for legislation, FINOS launched a Tier-1-backed fund with DTCC, Morgan Stanley, RBC and NatWest for open-source AI governance, and MAS published SAFR, built on the premise that agents now act faster than humans can intervene. 

Underneath it all, the BIS warned that the real AI bubble risk sits in opaque debt financing, not equity valuations, and a supply-chain-augmented research paper closes this week`s edition with a reminder that pricing information now moves through networks, not just filings.

AGENTIC COMMERCE, RAILS & INFRASTRUCTURE

1. Robinhood Starts Shipping Infrastructure While Positioining at the Intersection of the Agentic Economy and the Tokenized World

What Happened

Robinhood used its July 1 "The World Is Flat" keynote in London to tie three product moves into one architecture play. CEO Vlad Tenev and Johann Kerbrat announced Robinhood Chain going live on mainnet, an Arbitrum-based layer-2 for tokenized real-world assets, alongside round-the-clock Stock Tokens trading. 

The other Agentic signal: Robinhood's Trading MCP server, opened for US equities and options in May, is being extended to crypto through new Agentic Accounts, letting any AI model a trader chooses connect directly to Robinhood's data and execution stack. 

As a stunt, Robinhood also set a Guinness World Record for the most items an AI agent purchased in three minutes, using its Agentic Credit Card, certified on-site by an official adjudicator.

Why It Matters

This isn't the first brokerage to hand an AI agent the keys, Coinbase, eToro, Webull, Deriv, IG Group, and ThinkMarkets all shipped MCP-based agent trading. What sets Robinhood apart is scale and intent: nearly 28 million customers across 38 countries, and a stated goal, in Tenev's own words, of building a "financial superapp" where stocks, tokens, derivatives, prediction markets, and AI agents eventually sit inside one account. 

The Guinness stunt is testosterone marketing. Robinhood Chain and the tokenization push are the actual strategy, agentic trading is one more layer Robinhood intends to fold into a single integrated platform, not a standalone feature.

Robinhood's own disclosures don't guarantee the accuracy of agent-generated trades and place monitoring responsibility on the user, and MAS published a runtime safeguards framework this week addressing exactly that gap, Robinhood built the rail first and is writing the disclaimers as it scales.

DEPLOYMENT & USE CASES

2. Kueski's AI-First Underwriting Rebuilds Credit Access in Mexico

What Happened

Kueski's CTO Jaime Romero detailed how the Mexican fintech built real-time credit decisioning around alternative-data signals rather than adapting a legacy lending stack. The platform evaluates thousands of behavioral, transactional, device, and fraud signals simultaneously at checkout, combining underwriting and fraud prevention into one identity-and-repayment judgment rather than two sequential steps, with models retraining continuously against repayment outcomes. The results: over 40 million loans issued, Kueski Pay live with close to 40% of Mexico's leading e-commerce merchants, virtual assistants handling roughly 90% of customer interactions, and 20% of borrowers opening a bank account specifically to receive a Kueski loan.

Why It Matters

Kueski, founded in 2012 in Guadalajara, is Mexico's largest online consumer lender, built years before BNPL existed as a category. It has issued more than 30 million loans through two core products: Kueski Cash (personal loans) and Kueski Pay (BNPL), and has raised over $200 million in its largest funding round to date.

Building underwriting and fraud detection as one real-time decision, rather than sequential gates, is an architecture choice. Kueski built its risk infrastructure so fraud prevention, identity verification, and underwriting run together in real time rather than as sequential gates, a systems-integration choice, not a single model doing double duty.

The 20% figure, customers opening bank accounts to receive a loan, is the financial-inclusion proof point. This is what AI-native lending should mean in any market with thin-file borrowers, not just Mexico.

3. OCBC Bets AI-Native Wealth Advice and Headcount Growth on Each Other

What Happened

OCBC launched OCBC WoW, an AI-native wealth platform built around digital avatars positioned as round-the-clock advisers, for its Premier Private Client segment in Singapore, clients with at least S$1.5 million in assets under management. The bank will hire 600 relationship managers over three years alongside the platform and expects to spend more than S$1 billion annually on AI infrastructure. Multilingual support and expanded insurance and banking products are planned in later phases. Beta access is currently invite-only.

Why It Matters

OCBC is betting that AI-native wealth advice and headcount growth aren't substitutes. Framing the S$1 billion spend as growth and revenue enablement rather than cost-cutting is a deliberate signal to the market and to its own relationship managers, who might otherwise read "AI-native" as a threat to their jobs. Watch whether the AUM threshold moves down over time, that's the real test of whether this scales beyond ultra-high-net-worth clients.

MARKET IMPACT, FUNDING & INNOVATION

4. BIS: the AI Bubble Risk Isn't in Equities, It's in the Debt

What Happened

The Bank for International Settlements warned that debt-fuelled AI infrastructure spending is creating financial-stability risk through opaque financing links between hyperscalers, data-centre builders, and shadow banks. The concern is structural rather than about equity valuations: a sudden slowdown in hyperscaler capital expenditure could leave borrowers across that supply chain unable to service debt, and the financing itself is difficult to trace because it sits partly outside regulated banking. The BIS also flagged rising public debt, inflation pressure, and fragile sovereign bond markets, and called for greater transparency, targeted stress tests, and tighter oversight of leveraged AI investment.

Why It Matters

This shifts the AI-bubble conversation from the equity side, Nvidia multiples, hyperscaler valuations, to the debt side, where leverage actually breaks things. The AI capex meant to de-risk institutions through automation is itself becoming the systemic risk regulators worry about most. If your institution has any exposure to private credit funds financing data-centre buildouts, direct or through an asset-management partnership, this is the line for your next risk committee, sovereign risk and vendor concentration risk are now the same conversation.

REGULATION & GOVERNANCE

5. FCA and Bank of England: Agentic AI Is a "Profound Step Change," and They're Not Waiting for Legislation

What Happened

The FCA and Bank of England both described agentic AI as a "profound step change" for financial services in recent speeches, and signaled they intend to act ahead of formal legislation rather than wait for it. Priorities include board-level accountability for AI-driven decisions even when systems act autonomously, simulation and live testing to assess market-wide resilience and herding risk, and clearer frameworks for consent, authorization, and liability as agents transact on customers' behalf. Upcoming outputs, the Mills Review, an FPC update, and FCA Live Testing reports, are expected to translate this stance into supervisory practice.

Why It Matters

Two of the world's most consequential financial regulators just said, in effect, that static rulebooks won't hold against agentic systems acting faster than examination cycles can keep pace with. The emphasis on ecosystem collaboration, regulators, firms, vendors, and market infrastructures co-designing standards, is a tacit admission that no single regulator can write this rulebook alone. "Stewardship ahead of legislation" is the operative phrase, waiting for a finished AI Act analog before building governance is no longer defensible for a UK-regulated firm.

6. FINOS: Four Tier-1 Institutions Pool Resources for Open-Source AI Governance

What Happened

FINOS launched a member-led AI Fund backed by DTCC, Morgan Stanley, RBC, and NatWest, structured as a Supplemental Directed Fund with its own Governing Board, to pool capital and technical resources toward financial-grade governance for open-source AI. Initial priorities include advancing the FINOS AI Governance Framework, developing certification approaches, building interoperable specifications, and engaging regulators directly so supervisory expectations translate into implementable technical standards rather than abstract principles. The fund also plans community activation and training to reduce individual firms' dependency on proprietary governance tooling.

Why It Matters

This is the industry's answer to a question regulators keep asking and firms keep answering individually: what does financial-grade actually mean for open-source AI infrastructure? Four major institutions pooling resources rather than each building governance-as-code in-house is a bet that shared standards beat proprietary moats here, a notable contrast to how these firms compete on almost everything else. It also echoes Santander's open-sourcing of its AI governance stack, covered in the last edition, the direction of travel across Tier-1 institutions is toward shared, auditable governance infrastructure, not vendor lock-in.

FINOS is a nonprofit under the Linux Foundation where competing banks jointly fund open-source infrastructure. 

7. MAS, With Industry, Publishes SAFR: "AI Agents Now Act Faster Than Humans Can Directly Intervene"

What Happened

The Monetary Authority of Singapore, together with leading financial institutions and fintechs, published Safeguards for Agentic Finance at Runtime (SAFR), an industry-developed framework under MAS's BuildFin.ai initiative that builds on the earlier Project MindForge risk-management toolkit. MAS's rationale is direct: AI agents in financial services now act autonomously at a speed beyond practical human intervention. 

SAFR's four pillars are policy-bound execution, real-time validation, auditability, and interoperability. Singapore institutions are already applying it in wealth management, payments and treasury operations, and client engagement, where agents review documents, execute routine transactions within predefined terms, and draft client materials.

Why It Matters

SAFR is the most concrete governance artifact of the three regulatory items this week, a named framework with technical pillars and live production examples, not a speech or a discussion draft. That MAS built it jointly with industry rather than issuing it unilaterally matters too, a supervisory expectation with the institutions' fingerprints already on it tends to produce faster, less contested adoption than top-down rulemaking. "Faster than humans can intervene" is the sentence every board member outside Singapore should be quoting in their own governance discussions.

RESEARCH

8. Supply-Chain-Augmented LLM Embeddings Predict Which Firms the Market Hasn't Priced In Yet

What Happened

A new paper combines FinBERT embeddings of 10-K MD&A disclosures with supply-chain knowledge-graph propagation to test whether textual signals from one firm predict returns at connected firms. Using 255 S&P 500 firms from 2011 to 2025, the authors find a network-augmented factor with statistically significant cross-sectional return predictability that survives controls for momentum, volatility, and firm size. A long-short portfolio built on the factor generated an annualized Sharpe ratio of 0.86 and a Fama-French five-factor alpha of 7.27% per year, with robustness confirmed through out-of-sample tests, placebo experiments, and sector-neutral subsamples.

Why It Matters

This is a rare research item that clears the bar for both methodology and real financial-services application, quant and asset-management readers can act on it, not just cite it. The finding that pricing-relevant information propagates through supply-chain networks beyond what firm-level disclosures capture on their own has direct implications for anyone building factor models or NLP-driven signal generation. It's also a data point for the AI-induced unbundling thesis: information advantage is shifting toward whoever can model the network, not just whoever reads the filing fastest.

Source:

This newsletter is a human-AI collaboration involving seven AI agents orchestrated by Nicolas C. and my own curation and refinement.